Interest GSoC Project: Face-Detection with Runtime Device Switching

[Sachin-NK]

Dear Shivam Basia & Aishwarye Omer,

I am Sachintha Nadeeshan, a Computer Science undergraduate with experience in AI, cloud computing, and ML model deployment. I am particularly interested in contributing to the OpenVINO-based face-detection project for GSoC 2025, which focuses on runtime inference device switching.

To better understand the project, I have:

Explored OpenVINO’s AUTO feature and its approach to dynamic device selection.
Reviewed best practices for deploying face-detection models efficiently across NPUs, GPUs, and CPUs.
Investigated how model optimization impacts performance when switching between inference devices.
I would love to discuss:

• How OpenVINO ensures seamless transitions between devices without affecting inference quality.
• The key challenges in balancing latency, power efficiency, and performance during runtime switching.
• Any initial contributions I can make before GSoC begins.

Would it be possible to discuss this further via OpenVINO’s community forum or another platform? Looking forward to your guidance!

[omeraish]

Hello @Sachin-NK,

I apologize for replying late in the thread. I did not see this thread until today. Thank you for reaching out and showing your interest in the project.

The topics in the discussion are basically the ones we aim to answer and then demonstrate using the project. OpenVINO AUTO is an interesting feature which provides the capability to run inferences on a device without specifying it explicitly in the code. The idea of the project stems from above and we want to extend and demonstrate that inferencing happens seamlessly while switching devices during runtime.

You may refer to GFIs (Good first issue) section to see if any issue is of particular interest to you and work on it. Please refer to the homepage of the OpenVINO org to find more details on the prerequisite for your participation with the org.

We can use this thread to communicate for now. Also, if you have any past contributions to OpenVINO then let us know about it as well